jean-philippe vergne y rodolphe durand - theory and empirics path dependence

8/13/2019 Jean-Philippe Vergne y Rodolphe Durand - Theory and Empirics Path Dependence

1/25

The Missing Link Between the Theory and Empirics ofPath Dependence: Conceptual Clarification,Testability Issue, and Methodological Implicationsjoms_913 736..759

Jean-Philippe Vergne and Rodolphe Durand

HEC Paris

Path dependence is a central construct in organizational research, used to describe

a mechanism that connects the past and the future in an abstract way. However, across

institutional, technology, and strategy literatures, it remains unclear why path dependence

sometimes occurs and sometimes not, why it sometimes lead to inefficient outcomes and

sometimes not, how it differs from mere increasing returns, and how scholars can empirically

support their claims on path dependence. Hence, path dependence is not yet a theory since it

does not causally relate identified variables in a systematized manner. Instead, the existing

literature tends to conflate path dependence as a process (i.e. history unfolding in a

self-reinforcing manner) and as an outcome (i.e. a persisting state of the world with specificproperties, called lock-in). This paper contributes theoretically and methodologically to

tackling these issues by: (1) providing a formal definition of path dependence that disentangles

process and outcome, and identifies the necessary conditions for path dependence; (2)

distinguishing clearly between path dependence and other history matters kinds of

mechanisms; and (3) specifying the missing link between theoretical and empirical path

dependence. In particular, we suggest moving away from historical case studies of supposedly

path-dependent processes to focus on more controlled research designs such as simulations,

experiments, and counterfactual investigation.

INTRODUCTION

Broadly speaking, path dependence is about increasingly constrained processes that

cannot easily be escaped. As already noted by Sydow et al. (2009), path dependence

has become an essential theoretical construct for many organization scholars. A data-

base search in seven leading organization and management journals for the period

19982007 shows that the notion of path dependence becomes more and more

popular. Between 1998 and 2002, 109 articles published in Academy of Management

Journal, Academy of Management Review,Administrative Science Quarterly, Journal of Management

Address for reprints: Jean-Philippe Vergne, HEC Paris, Business Policy and Strategy Department, 1 rue de laLibration, 78 351 Jouy-en-Josas Cedex, France ([email protected]).

Journal of Management Studies47:4 June 2010doi: 10.1111/j.1467-6486.2010.00913.x


2/25

Studies, Organization Science, Organization Studies, and Strategic Management Journal make a

direct reference to path dependence, path dependency, or path-dependent processes.

This accounts for 6.15 per cent of all papers published in these journals. Between 2003

and 2007, the figure increases to 214 articles, i.e. 10.5 per cent of all articles published.

Despite the common use of the construct in scientific publications, a number of schol-

ars note that no clear definition of path dependence is shared within the scholarly

community (Djelic and Quack, 2007; Morgan and Kubo, 2005; Pierson, 2000).

Indeed, it is hard to tell what constitutes acceptable empirical evidence for path depen-

dence. Some scholars remain sceptical about the empirical support for path depen-

dence (Liebowitz and Margolis, 1990), while others go as far as to wonder whether

path dependence operates in the real world . . . or whether it is purely a theoretical

artifact (Hirsch and Gillespie, 2001, p. 72; see also Ekelund and Tollison, 1997,

p. 387; Liebowitz and Margolis, 1994).

We believe it is time to take stock of past research and analyse carefully the theoretical

and empirical underpinnings of path dependence. It should appear problematic to thescholarly community that one published article in ten refers to a blurry and controversial

construct. Thus, this paper proposes to critically review the use of path dependence at

three levels of analysis. At the macro level, institutionalists use path dependence to

account for (harmful) institutional persistence (Djelic and Quack, 2007; Mahoney, 2000;

Morgan and Kubo, 2005; North, 1990; Pierson, 2000; Thelen, 1999). At the meso level,

economists rely on path dependence to explain suboptimal governance or technology

outcomes (Arthur, 1989, 1990; Cowan, 1990; David, 1985; Gedajlovic et al., 2004;

Williamson, 1999). At the micro level, the dynamic capability view refers to path

dependence as a surrogate for organizational rigidity while paradoxically insisting on itspositive impact on competitive advantage (Eisenhardt and Martin, 2000; Helfat and

Peteraf, 2003; Schreygg and Kliesch-Eberl, 2007; Sydow et al., 2009; Teece, 2007;

Teece et al., 1994, 1997; Vergne and Durand, 2010).

In contrast with past organizational research, we offer a narrow definition of path

dependence as a property* of a stochastic* process which obtains under two conditions

(contingency* and self-reinforcement*) and causes lock-in* in the absence of exogenous

shock.[1] According to this definition, path dependence ceases to be a ubiquitous con-

struct describing most historical sequences. Instead, we privilege explanatory power: our

definition of path dependence is thus applicable to a restricted set of sequences, but with

a greater theoretical and empirical value. We take a step further by discussing the missing

link between the theory and empirics of path dependence. Important issues have pre-

vented scholars from testing path dependence empirically in an adequate way. Our

redefinition emphasizes the two conditions necessary to obtain path dependence, and to

clearly distinguish it from other history matters kinds of theoretical constructs. Because

a necessary condition* of path dependence, namely contingency, is practically impossible

to verify* or falsify* in case study research (Bassanini and Dosi, 2001; David, 1985;

Foray, 1997), and because misleading claims about suboptimality* threaten the validity

of path dependence research (Goldstone, 1998; Hirsch and Gillespie, 2001; Liebowitz

and Margolis, 1990, 1995), we recommend using highly-controlled methodologies in thefuture to develop more robust path dependence research (i.e. simulation, experiment,

Path Dependence 737


3/25

PATH DEPENDENCE AT THE MACRO, MESO, AND MICRO LEVELS

Past research in organization science demonstrates that we can do without the notion of

path dependence to understand that history matters. For example, we know that what

managers learn today influences what they will be able to learn tomorrow as absorptive

capacity increases (Cohen and Levinthal, 1990). Where investments are made now islikely to constrain where they can be made in the future (Dierickx and Cool, 1989).

Under certain conditions, firms that move first will gain a sustained advantage over their

competitors (Lieberman and Montgomery, 1988). Institutions tend to evolve incremen-

tally rather than radically, so yesterdays rules of the game are often very similar to

todays rules (Scott, 1995; Selznick, 1957). Organizational characteristics tend to reflect

how things were at the time of organizational founding (Marquis, 2003; Stinchcombe,

1965). Most organizations display inert structures, as if their past history determined

their present state (Hannan and Freeman, 1984). Absorptive capacity, first-mover advan-

tage, institutional persistence, imprinting, or structural inertia are all well-known theo-

retical mechanisms that explain how certain aspects of the past relate to current

properties of organizations. Path dependence, too, proposes to connect the past and the

present in a formal way, at the (macro) level of institutions, at the (meso) level of

technology and governance modes, and at the (micro) level of organizational resources

and capabilities.

The Macro Level: Institutions

Path dependence scholars view institutions as carriers of history which maintain existing

behavioural norms and cultural patterns throughout time (David, 1994). From an insti-tutional perspective, the interest of path dependence seems to reside in the potential

inefficiency of history (March and Olsen, 1989) that results from the stickiness of

institutions. North maintains that once an economy is on an inefficient path . . . it can

persist because of the nature of path dependence (North, 1993, p. 3). For example,

organizations that strive within a given institutional matrix have a stake in perpetuating

the rules of the game that favour their own survival, even when such rules are globally

inefficient, thereby hampering institutional change (North, 1990; Pierson, 2000).

Although institutional scholars traditionally associate path dependence with an absence

of change leading to persisting inefficiencies, a more recent view shifts the focus from

unproductive status quo to novelty. As Garud and Karne put it, novelty. . . is not a

negation of the past, but an elaboration and extension in specific directions depending on

the particular sequence of unfolding events. Stated differently, the emergence of novelty

is a path dependent phenomenon (Garud and Karne, 2001a, p. 1). While this definition

enriches the institutional perspective on entrepreneurship (e.g. Garud, 2008), it also

broadens the list of possible outcomes of path dependence to encompass innovation in

addition to persistence (see also Garud et al., 2010; Porac et al., 2001). Surprisingly, the

same mechanism path dependence could explain either the persistence of existing

(and possibly inefficient) institutions, or the creation of new ones. Innovation is certainly

about recombining previous knowledge (Kogut and Zander, 1992), and is thus past-dependent, yet path dependence goes beyond the mere idea that the past channels future

J.-P. Vergne and R. Durand738


4/25

Furthermore, if we define institutions as stable social patterns and identify path

dependence with persistence, as is usually the case, then is it not tautological to speak of

path-dependent institutions? If not, what does path dependence add to the standard

characterization of institutions? A way to clarify the debate is to clearly distinguish

between the process of path dependence (which, we will show, consists of contingency

and self-reinforcement) and its outcome (a specific state of persistence called lock-in).

This distinction is helpful because it makes visible the divide between path dependence

scholars who accept the contingency assumption and those who doubt its relevance. For

example, several researchers (Garud and Karne, 2001b; Rao and Singh, 2001; Sydow

et al., 2009) reject the idea that novel paths emerge serendipitously, that is, they deny the

role of accidental events in path origination to focus instead on human agency, while

others insist that small contingent events are necessary to distinguish path dependence

from the mere action of institutional increasing returns*, sunk costs, or adaptive expec-

tations (Arthur, 1989; David, 1985; Goldstone, 1998; North, 1990; Schwartz, 2004).

Accordingly, such mechanisms would contribute to path dependence only as amplifiersof small events and chance circumstances, identified as the true origin of paths (North,

1990, p. 94; see also Arthur, 1989; David, 1985).

Thus, the usefulness of path dependence is bound to remain limited if it explains both

persistence and its opposite (i.e. novelty); path dependence may even be tautological, or

superfluous relative to existing theories, as we will further discuss in the following.

The Meso Level: Technology and Governance

At the meso-level, the definition of path dependence suffers from the same ambiguities.Path dependence often appears to be a quasi-synonym of persistence. Helfat (1994, p.

1722) sees persistence in R&D as an example of a wider phenomenon termed path

dependence . Cowan and Gunby (1996, p. 521) speak of technological inertia caused

by increasing returns to explain why certain paths have a tendency to become

entrenched. In their case study of pest control strategies, the two authors maintain

that path dependence induces persisting inefficiencies at the technology level. Similar

claims, all based on historical case studies, have been made about the technological

development of keyboards (David, 1985), nuclear power reactors (Cowan, 1990), or

videocassette recorders (Arthur, 1990). The overarching logics behind technological

path dependence is grounded in the fact that repeated R&D investment creates sunk

costs, which result in irreversibilities along the path of technology development (David,

1985; Helfat, 1994).

At the level of organizational governance, path dependence is also recognized as a

serious issue which can account for persisting inferior governance structures (Williamson,

1999). For example, Gedajlovic et al. (2004, p. 900) argue that path dependencies

. . . inhibit the transformation of founder- into professional-managed firms. Repeated

patterns of investment in human or material resources lead to routine creation and asset

specificity, which both introduce stickiness at the governance level and prevent subse-

quent adjustment (Gedajlovic et al., 2004; Williamson, 1999).Broadly speaking, when it comes to technology development or governance structures,

Path Dependence 739


5/25

capital or asset accumulation. As was already the case in the institutional literature, the

outcome of such processes can, but need not lead to suboptimal outcomes (Cowan and

Gunby, 1996; David, 1994). The existing literature gives little guidance to understand

what determines the nature of outcomes (e.g. persistence or change) and their properties

(e.g. suboptimal or not). Liebowitz and Margolis (1995) remark that if path dependence

simply means that investment decisions sometimes appear inefficient ex postwhen new

information is revealed, then path dependence is nothing but an intertemporal propa-

gation of error caused by information incompleteness. That is, the notion of incomplete

information better captures temporary irreversibility in investment sequences than path

dependence does. The persisting indeterminacy in the conceptualization of path depen-

dence raises the issue of whether the construct is theoretically distinctive, and what it

adds as a conceptual mechanism explaining institutional, technological, and strategy

persistence.

The Micro Level: Firm Resource and Capability

At the resource and capability level, path dependence is a central concept to the

resource-based view (RBV) and the dynamic capability view (DCV) (Eisenhardt and

Martin, 2000; Helfat and Peteraf, 2003; Schreygg and Kliesch-Eberl, 2007; Teece,

2007; Teece et al., 1997). Strategic management scholars associate path dependence

with the kind of persistence caused by repeated investment in firm resources and

capabilities, which can create competitive advantage (Barney, 1991; Dierickx and Cool,

1989; Teece, 2007) but also rigidities and harmful lock-in (Leonard-Barton, 1992; Teece

et al., 1997; Tripsas, 1997). Specifically, the outcome of path-dependent capabilitydevelopment is more likely to be positive for the firm when asset complementarities,

learning specialization, or increasing returns to scale and scope prevent imitation by its

competitors. As Helfat and colleagues put it, the idea that capabilities emerge from a

series of path-dependent learning experiences is a central one. Further, competitive

advantage requires that these capabilities are valued by customers, and are difficult to

imitate by competitors (Helfat et al., 2007, pp. 623).

Paradoxically, while capability development is path-dependent, the role of dynamic

capabilities is to purposefully create, extend, or modify [the firms] resource base (Helfat

et al., 2007, p. 1) to increase environmental fitness and avoid lock-in. At a theoretical

level, dynamic capabilities emphasize the managerial ability to orchestrate a firms assets

so as to avoid harmful inertia in a rapidly changing environment (Teece et al., 1997). It

follows that dynamic capabilities are construed simultaneously as path-dependent and as

a remedy to path dependence. Because of this ambiguity, the capability lifecycle frame-

work describes path-dependent capability trajectories but fails to specify when path

dependence will lead to capability retirement, replication, or recombination (Helfat and

Peteraf, 2003).

At the micro-level, it remains unclear how exactly path dependence differs from

resource accumulation or capability building (i.e. persistence of a micro-organizational

pattern). Again, the definition appears too broad because it does not sufficiently exploittwo distinctive conditions for path dependence: contingency and self-reinforcement.



6/25

may cause lock-in on inferior technologies, and that explain why companies with

the best products will not always win (Teece et al., 1997, p. 523)? If we agree that most

investments in resources and capabilities are carefully thought through, then we should

be able to propose a distinctive definition of path dependence that attributes correctly

to a given outcome (success or failure) its antecedents by reconsidering the causal

powers of chance events relative to resource/capability properties. By the same token,

we should be able to dissociate accumulation from other self-reinforcing mechanisms to

better define path dependence and avoid theoretical confusion (Vergne and Durand,

2010).

WHAT IS PATH DEPENDENCE, REALLY, AND (WHY) DOES

IT MATTER?

We define path dependence as a property of a stochastic process which obtains under

two conditions (contingency and self-reinforcement) and causes lock-in in the absenceof exogenous shock. Because it has formal theoretical roots in the mathematics of

Markov chains (see Appendix 1), path dependence should not become an umbrella

category for all theories that explain why institutions, technological standards, or firm

capabilities tend to persist over time. Table I presents several alternative theories

according to which history matters, and shows how they differ from path dependence

along several features (initial conditions, triggering event, sustaining mechanism, level

of analysis, outcome, and degree of outcome probability). There lies the theoretical

interest of path dependence: if we accept to narrow down its definition to its logical core, and

recognize that not every historical process is path-dependent, it can explain phenomena that othertheories cannot.

History Unfolding: Two Conditions for Path Dependence

Condition 1. Contingency. What triggers path dependence is still unclear in the literature.

While some scholars associate path dependence with sensitive dependence on initial

conditions (Baum and Silverman, 2001; Liebowitz and Margolis, 1995), Goldstone

(1998, p. 834) maintains that path-dependent outcomes are not determinedby any par-

ticular set of initial conditions. In fact, path dependence occurs when initial conditions

are followed by a series of contingent (or chance) events whose influence on the path

taken is larger than that of the initial conditions themselves. Consistent with prior works,

by contingent we mean unpredictable, non-purposive, and somewhat random events

(Arthur, 1989; De Rond and Thietart, 2007). For instance, a first mover advantage is

rarely contingent: firms strategically plan to enter first in markets when they know it can

be beneficial. Therefore, a persisting market domination explained by first mover advan-

tage would not qualify for a path dependence explanation.[2]

The existence of contingency in organizational life is a reasonable assumption.

Kenney and von Burg describe the good fortune at play in luring [William] Shockley

back to Palo Alto in the early 1950s (Kenney and von Burg, 2001, pp. 1378). Theco-inventor of the transistor could not find investors on the East Coast, where potential

Path Dependence 741


7/25

ableI.

Pathdependenceandotherhistorymattersnotions

heoretical

gument

Influenceofinitial

conditions

Triggeringevent

Sustainingmechanism

Levelofanalysis

Outcome

Degreeofoutcome

predictability

Representativeworks

athdep

endenceVeryweak

Co

ntingent

events

Self-reinforcement

Multiple

(scalability)

Lock-in

Verylowexante;

mediumasself-

reinforcementstarts

A

rthur(1989);

D

avid(2001)

bsorptive

apacity

Weak

Knowledge

sto

ck

Knowledgeflows

Group:

organization

Innovation,

perform

ance

Medium(depends

onmarketdemand

andcompetition)

C

ohenand

L

evinthal(1990)

nstitutio

nal

ersistence

Strong

N/A

Stickinessof

institutional

patternsata

socio-cognitive

level;isomorphism

Institution

Institutional

stability;

increme

ntal

change

Importantin

absenceof

exogenousshocks

N

orth(1990);

Selznick(1949);

Scott(1991)

esource

ccumulation

Strong(managerial

commitment)

Managerial

decision

Repeated

commitment

Resource,

capability

Uniqueness,

non-replicability

High

H

elfat(1994);

D

ierickxandCool

(1

989)

ructuralinertiaStrong

Firmfounding

Accountability,

reliability

Environment/

organization

Increase

dfirm

survival

(earlyin

theproc

ess)

Mediumtohigh

(dependsonfirm

environment)

H

annanand

Freeman(1984)

mprinting

Verystrong

Dateandplace

of

founding

Institutionalization

Organization

Persistenceof

firmstru

ctural

properti

es

Mediumtohigh

(dependsonfirm

environment)

Stinchcombe

(1

965);Marquis

(2

003)

rst-mover

dvantag

e

Verystrong

Orderofentry

Entrybarriers,

pre-emptionof

scarceresources

Organization-marketMarket

predominance

Lowtomedium

(dependsonrivals,

industrystructure

andregulation)

L

iebermanand

M

ontgomery

(1

988)

haosth

eory

Maximal

N/A

Hyper-determinism

+

existenceof

chaoticattractor

Multiple

(scalability)

Entirely

determinedby

initialconditions

Verylow

B

aumand

Silverman(2001);

M

cKelvey(1999)



8/25

go back to California. There he trained eight engineers who eventually left his lab to

create Fairchild Semiconductors, a company which catalysed the pattern of new firm

formation that put the silicon in Silicon Valley. Contingency in organizational life can

take many shapes (e.g. unexpected encounters, trial-and-errors leading to unattended

consequences) and seems to have potential for long-term effects.

Condition 2. Self-reinforcement. The unpredictability inherent to path dependence comes

from the fact that, as the process unfolds, the outcome distribution changes that is,

the likelihood of obtaining a certain outcome varies with time. This is the case because

path dependence is a property of processes similar to non-ergodic* Markov chains.

Once a path has been contingently selected, various mechanisms can lead to its

self-reinforcement, such as positive network externalities* or increasing returns (e.g.

to scale, to scope, to learning) (Arthur, 1989; Bassanini and Dosi, 2001; Pierson,

2000). For example, QWERTY keyboard production becomes more and more

profitable as production scale increases, because fixed costs are distributed across alarger number of units. From the users perspective, getting accustomed to QWERTY

keyboards makes typing easier and quicker, so that the more a user types on

QWERTY, the more utility he/she derives from it. Production scale and user learning

loops both contribute to reinforcing QWERTY as the preferred keyboard. More gen-

erally, features of self-reinforcement are very common in organizational life (Arthur,

1989).

Page (2006, p. 90) establishes that increasing returns are neither necessary nor

sufficient for path dependence: what is really required to sustain a path is a mechanism

that decreases the relative attractiveness of alternatives (Arrow, 2000; Kay, 2005).Imagine that agents have to choose between only two possible paths which both yield

decreasing returns; logically, lock-in should occur on the path whose returns decrease at

the lowest speed. Learning loops for QWERTY users not only make QWERTY more

attractive, but also alternative keyboards less attractive, because of the time it would take

to reach the typing performance of an advanced user on a new system. Thus self-

reinforcement, to be effective, needs to include at least one negative externality to

decrease the attractiveness of alternative paths.

Path Outcome: Lock-In

When a process possesses the property of path dependence, then lock-in will occur on

one of the possible outcomes if no exogenous shock disturbs the system. Lock-in char-

acterizes a state of equilibrium with a very low potential for endogenous change put

simply, lock-in is a hard-to-escape situation. Path dependence has a true theoretical

substance that basically says: when contingently selected paths undergo self-

reinforcement, then, ceteris paribus, lock-in will occur on one single path as alternative

options are selected out (owing to negative externalities). Lock-in is a state of the system

that cannot be escaped endogenously. Because paths are selected contingently, lock-in

can happen on any path, i.e. not necessarily on the optimal one. Put differently, initialconditions do not determine which equilibrium will eventually prevail,[3] and path

Path Dependence 743


9/25

nature of the underlying process. Formally, path-dependent processes are such that, for

any set of initial conditions I, the probability of any outcome O conditioned by I verifies:

"(O,I), P(O|I) < 1. This theorem states that it is impossible to foresee with certainty the

outcome of a path-dependent process when only its initial conditions are known.

The QWERTY case is a well-known example of alleged lock-in on a suboptimal*

path. Arguably, tremendous exogenous influence would be required for the QWERTY

standard to be replaced by another one (David, 1985; for a critique, see Liebowitz and

Margolis, 1990). If it is correct that QWERTY was chosen contingently among alterna-

tive keyboards and is not the best keyboard, then the history of keyboard standards must

have been path-dependent.

Empirical Scope and Theoretical Interest

At this stage of the paper, we have proposed a clear definition of path dependence,

detailing how paths are created, how they reinforce themselves, and how they reach astate of equilibrium in the absence of exogenous shocks. While all necessary components

of path dependence, taken individually, are common in organizational life, their

adequate combination might be less common, so it remains unclear how often path

dependence is likely to manifest itself empirically. Are stories of the QWERTY type

exceptional, or do they represent a significant proportion of technological paths? Cross-

national comparison in a given industry often provides interesting clues. For example,

the fact that combustion engines independently emerged as a standard in all developed

countries despite very different initial conditions (Foray, 1997; Foreman-Peck, 1996)

should rule out, as possible explanations, both chaos theory (e.g. no hypersensitivity toinitial condition) and path dependence (e.g. no outcome diversity). By contrast, when we

observe outcome diversity despite similar initial conditions, path dependence is a serious

candidate to account for differential evolutionary paths. At the institutional level, Bas-

sanini and Dosi (2001) note that Cipollas model of Europes development in the modern

age gives a central role to minor lucky events with huge intertemporal consequences on

the worlds balance of power (Cipolla, 1965). At the technology level, Schreygg and

Blinn (2008) remark that dubbing in the film industry has persisted in several countries

with a small cinema audience despite its large fixed cost and the existence of a better

option (i.e. subtitling). At the resource and capability level, only Pfizer was lucky enough

both to remark that antihypertensive drugs had side-effects on male erection and to seize

the opportunity to pursue a new research path following a contingent discovery (De

Rond and Thietart, 2007).[4]

LOOKING FOR THE MISSING LINK: THE EMPIRICAL TESTABILITY

OF PATH DEPENDENCE

As noted by Hirsch and Gillespie (2001, p. 43), the proposition that history matters

intuitively goes together with some sort of thought experiment (or counterfactual) which

can be rarely undertaken through an actual experiment (see also Durand and Vaara,2009). Because testing path dependence involves a comparison between the current



10/25

followed (Cowan and Gunby, 1996, p. 521), it is still contested whether case studies such

as QWERTY or VHS constitute acceptable empirical evidence (Liebowitz and

Margolis, 1990, 1995). Drawing on epistemology, this section addresses three empirical

issues that have so far precluded path dependence research to produce undisputable

empirical findings. In each case, we unpack the conditions under which observable

phenomena can be said to verify or falsify path dependence.

Issue 1: The Contingency Assumption Verifiability

For the sake of simplicity, and consistent with past research (Arthur, 1989; Kay, 2005),

the following discussion models path-dependent processes as adoption patterns in the

broad sense: social actors adopt technologies, institutions, or develop resources, and

some of these adoption paths end up dominating others (e.g. lock-in occurs). Adoption

patterns can be represented as sequences of digits: for instance, when two entities A and

B compete for adoption, the sequence 01101100110111100010 means that B waschosen first, then A twice, B again, etc., with 0 and 1 coding, respectively, for adoptions

of B and A. Consistent with the assumption that contingent events appear random to us

(Arthur, 1989; De Rond and Thietart, 2007), contingency is modelled hereafter as

random events. We recur to an intuitive approach of randomness derived from the works

of mathematicians (Chaitin, 1975; Li and Vitanyi, 1997; Solomonoff, 1960), according to

which a series of events appears more random than others when more information is

necessary to generate it (Chaitin, 1975). For instance, using computer language, the

program Write 1 20 times yields a series of digits that appears less random than the

output given by the instruction Write 01101100110111100010, for it uses a smalleralgorithm (i.e. 18 characters instead of 26). This is consistent with the idea that random

information is information that cannot be reduced to something simpler. This is why a

human brain usually estimates that 11111111111111111111 appears less random than

01101100110111100010.

In his adoption model, Arthur (1989) identifies the source of lock-in on technology B

in the early contingent lead in Bs adoptions, later reinforced by increasing returns.

According to Arthur, contingency means that early in the adoption process, something

like twenty adoptions of B in a row occurs and gives B a definitive advantage over A.

Such a statement is counter-intuitive though, since repetitive patterns are more likely to

have a non-random origin, as explained above. Yet Arthur considers that such a repeti-

tion is the true chance event that triggers path dependence.

By contrast, as organization scientists, we tend to believe that the more repeated a

sequence of events in a bounded time period (e.g. organizational failure, CEO dismissal),

the more likely a structural explanation can account for it. This is indeed a crucial

assumption on which much social science research is based. Consequently, when path

dependence is invoked to explain why VHS won over Betamax on the ground that an

initial lead in VHS adoptions was subsequently amplified by positive network externali-

ties, it is essential to justify that the initial lead was contingent and not caused by

structural forces. Alternatively, if we believe that VHS gained an initial advantage overBetamax because the former offered higher customer value due to longer recording time

Path Dependence 745


11/25

VHS machines than Beta producers could make early on in the competition (Cusumano

et al., 1992, p. 47), we should acknowledge that VHSs success is trivially explained by

product quality and strategic savvy. Seen from this perspective, path dependence looks

like a superfluous theoretical subtlety hence the controversy.

To claim that something is contingent is not quite like proving it actually is. What kind

of random events (i.e. not attributable to VHSs superiority as a competitor) could

account for VHSs initial lead? Chaitin (1975) asserts that evidence for randomness

cannot easily be produced. Going back to our simple model of adoption patterns, to show

that a series of digits is random, one needs to prove that no program of a complexity

smaller than that of the series exists that could generate it. According to Chaitin, Gdels

incompleteness theorem contains the rationale for why such evidence is often impossible

to produce. Any formal system consisting of a formal language and a set of inference rules

can be associated with a given degree of complexity. The more complex the system, the

more complex the information derivable from it (e.g. theorems and their proofs). As a

consequence, complex proofs cannot be established within simple systems: the informa-tion they contain is considered uncompressible and therefore appears random. This

inherent limitation of formal systems leaves theoretical systems like those employed in the

social sciences in an even more difficult position for proving randomness or chance. The

bottom line is: organization scholars will not be able to verify path dependence empiri-

cally if their argument relies on the ex postdemonstration that something did happen

contingently.[5]

The chancy character of an adoption pattern is unlikely to ever be empirically

evidenced. While chance remains a reasonable assumption in formal models where A

and B only differ in their payoff structure (e.g. Arthur, 1989), in real-world situations likeVHS vs. Betamax, customers adopt A or B based on their preferences and the perceived

properties of A and B. A causal link certainly exists between the idiosyncratic properties

of a technology (e.g. design, technical specifications, brand reputation) and the decision

to adopt it. Thus in empirical studies, the chance argument is no more provable than in

formal models, but it is even less realistic since human agents are not assumed to behave

randomly. In other words, the verifiability of contingency, a necessary condition for path

dependence, remains a serious issue.

Issue 2: The Contingency Assumption Falsifiability

One may wonder how it can be argued that VHSs early advantage is due tosomething else

than chance, that is, how the contingency assumption can be falsified. Suppose that A

and B stand for two entities competing for adoption, and whose payoff structures are

hard to compare early in the process. In order to disconfirm empirically the path-

dependent explanation based on contingency, one needs to show that B possesses at least

one property that differs from As which could explain why B got adopted. A and B

certainly differ in at least one of their respective properties: if they did not, then A and

B would not be distinct.[6]

To falsify contingency, it is necessary to assume that there exist property differencesbetween A and B that could capture variance in adoption trends. Yet, relating both



12/25

challenge: extensive data about the adopters and promoters motivations needs to be

gathered and analysed. A way to do so is to regress the speed of adoption on every

relevant property difference perceived by adopters. Had A and B exactly the same

properties but design, one could still argue that B was more fashionable, existed in a

wider variety of colours, or was more ergonomic, to explain why it got ahead. Any

combination of property differences could also be investigated as a cause for early

adoption; this means considering potential interaction effects between the independent

variables of the regression model. As a matter of fact, a lot of alternative hypotheses can

be formulated to account for Bs early advantage. Nevertheless, the quantity of infor-

mation required to test them is likely to exceed by far the quantity of information

available in existing historical accounts about the case under scrutiny. The VHS con-

troversy provides a good illustration of the argument. Arthur (1990, p. 92) describes the

similarity of VHS and Betamax along three properties: price, time of market entry, and

initial market shares. He infers that the final victory of VHS can only be explained by a

contingent lead amplified by increasing returns. Liebowitz and Margolis (1995, p. 222)contest Arthurs example by considering a fourth property that could account for the

domination of VHS, namely its longer recording time a property particularly valued by

consumers in the late 1970s.

Many non-contingent properties can explain the outcome of adoption patterns, yet it

is practically impossible to isolate structural causes when we observe a unique historical

trajectory. Therefore, the contingency assumption of path dependence is falsifiable only

discursively in historical case studies: while it is feasible to imagine how things could have

unfolded differently, it remains impossible to explain scientifically why they happened

the way they did. Hence, to rely on contingency is to take the risk of constructing atheoretical edifice on disputable, shaky foundations.

Issue 3: Long Run Suboptimality

Another issue that severely constrains the empirical validity of path-dependent expla-

nation is the idea that history matters in the long run. In the long run, VHS will be

replaced by another recording system (e.g. DVD) and QWERTY keyboards substituted

for something else (e.g. voice recognition systems). In the absence of exogenous shock,

the long run signals the moment when lock-in can empirically be observed. Therefore,

the lock-in phenomenon and its timing can be known only ex post. Since path depen-

dence predicts lock-in but does not tell when it will occur, the proposition that path

dependence implies lock-in suffers from a lack of falsifiability. A simple way to falsify

path dependence would be to observe that lock-in has not occurred; but since it cannot

be known when it is supposed to occur, one can only acknowledge that lock-in has not

occurredyet. Thus, the path dependence explanation cannot be falsified because it does

not specify a time limit after which the predicted observation (i.e. lock-in) can be said

not to have occurred.

Therefore, empirical evidence based on ex post case study observations is bound to

remain inconclusive (Liebowitz and Margolis, 1995) because it appeals to the observerssubjective perception of what is the long run equilibrium. It follows that one can easily

Path Dependence 747


13/25

termed the long run: for instance, one could maintain that history did not matter in the

VHS vs. Betamax competition since nowadays everyone is locked in using DVDs. To

falsify this claim, one needs to prove that a standard inferior to DVD would dominate

todays industry had Betamax cornered the market in the 1990s. As we will discuss

below, such proof is challenging to provide. Another implication of adopting a subjective

viewpoint is that optimality becomes relative to a particular audience. Strategically, it is

often optimal for a firm to impose its own technological standard, independent of the

inherent welfare implications of the standard at a societal level. What is strategically

optimal for a firm is not necessarily optimal for the whole society.

The difficulty of applying the theory of path dependence to historical observations

partly comes from the fact that, while some mathematical processes do converge to an

uncontroversially stable equilibrium, real history does not. Claims that our world is

suboptimal are not unlike Leibnizs attempt, more than three centuries ago, to establish

logically that we live in the best of all possible worlds (Leibniz, 1686/2000). Path

dependence scholars, with their emphasis on suboptimality, maintain that we donotlivein such an optimal world (David, 1985), because had the past been different, we could all

be better off right now (e.g. nobody would use inferior QWERTY keyboards). The

theorem of path dependence asserts that the past shapes the present stochastically:

"(O, I), Pr(O|I)


14/25

outcome did not occur, for contingency or self-reinforcement or both can be involved

in the non-selection of a given path.Ex post, we are not in a position to identify: (i) what

should have been different in the past to have a different future; and (ii) whether human

agency could have made that critical course of events different or not.

To illustrate this idea, let us imagine a counterfactual world wherein Betamax won

over VHS. In our actual world, the cash generated by VHSs success was reinvested to

further improve image quality, and led to the design of the superior S-VHS standard. But

in the counterfactual world, Sonys cashflows were entirely redistributed to shareholders,

and the Betamax standard never improved enough so as to match the S-VHS technol-

ogy. Worse, consumers got used to the idea that Betamax had reached the highest video

quality possible, and no firm ever saw the market opportunity to develop a better

standard, such as DVD.

This thought experiment helps us to understand the vacuity of hypothetical regret (i.e.

if only the past had been different, then . . .) in a path-dependent world. When one

maintains that QWERTY is suboptimal and that Betamax would have led to a betterfuture, one implicitly contradicts a core theorem of path dependence, i.e.

"(O, I), Pr(O|I)


15/25

simulations present obvious advantages that enable researchers to define initial condi-

tions, model contingency, and determine self-reinforcing mechanisms. A computer

simulation can take a complex set of assumptions, simulate a set of organizational

processes, and represent the implications of these processes for organizational outcomes

(Lant and Mezias, 1990, p. 151). Simulations force researchers to make explicit assump-

tions about a finite amount of known variables that can be controlled for. Researchers

are thus able to generate multiple historical trajectories emanating from the same set of

initial conditions, thus enabling them to generalize about the mechanisms and processes

that produce such histories (a case study, in contrast, must rely on the observation of one

historical path) (Zott, 2003, p. 109). Simulations can help assess the probability of path

dependence given a set of initial conditions, a certain amount of contingency, and the

nature of self-reinforcement, thereby providing scholars with estimates of the probable

frequency of path dependence in organizational life. For instance, Zott (2003) questions

the conventional wisdom on path dependence and dynamic capabilities and demon-

strates in a simulation study that path dependence is not a necessary property ofcapability development. Simulations have the potential to record the detailed charac-

teristics of unfolding processes. They can be re-run and their results compared since

complete data series are available for each relevant variable. It then becomes possible,

unlike inex posthistorical accounts, to trace the complex mechanisms that led (or not) to

lock-in.

Moreover, simulations display high internal validity and are particularly well suited

to study longitudinal process phenomena (Carley, 2001; Davis et al., 2007). When

looking at complex phenomena over time, simulations help answer the question What

if? rather than What happened? (Dooley, 2002), which makes them especially inter-esting for disentangling the intricacies of contingency. In this vein, Sterman and Wit-

tenberg (1999) simulate the path-dependent emergence of scientific paradigms and

observe that it is largely driven by situational characteristics of the environment. For

instance, they find that the confidence level that other paths represent serious alterna-

tives decreases the probability of lock-in. In their simulation of path-dependent orga-

nizational search processes, Becker et al. (2008) observe that organizational exploration

and coordination mitigate the potentially harmful effects of path dependence. Akker-

mans and Romme (2008) simulate the path-dependent evolution of a supplier network

and find that initial conditions inner dispositions towards network partnership have

little predictive value for the actual behaviour of network agents (for another simulation

of search paths controlling for the inner dispositions of search agents, see Winter et al.,

2007). In their simulation consistent with the theorem of path dependence (i.e.

"(O, I), Pr(O|I)


16/25

of a finite amount of variables, which renders falsification procedures feasible. Like in

simulations, choices are not assumedex postto be contingent, but are actually modelled

as the consequence of a random sampling procedure. While in a simulation the impact

of chance can be estimated by re-running the simulation with a different random seed,

in a lab context the experiment can be replicated with different, randomly-chosen

subjects. Using such designs, scholars can tell what is attributable to chance and what is

not, since all relevant variables are tracked at each step of the research process. In a

controlled experimental study of decision processes, Koch et al. (2008) find that path

dependency results from poor decisions that are due to the fact that people in highly

complex situations tend to neglect future developments at the expense of information on

present situations. Arguably, this finding can only appear when different

experimentally-controlled paths are compared along the dimension of decision complex-

ity. Importantly, since simulation and experimental parameters are set ex ante by the

researcher, the time boundaries of these studies are not subject to ad hocdetermination

based on ex postobservations. Parameters can be subsequently modified in replicationstudies to actually compare various regions of equilibria, which provides sufficient

support for claims about suboptimal paths. Experimental design can be used to study

controlled property differences. For example, Bach (2008) accounts for which property

differences between cable and DSL explain adoption in the German broadband market.

Looking at speed, price, contract duration, and technology, findings show that to com-

pensate for an initial lead in DSL adoption, cable operators need to propose contracts

whose perceived utility is at least 18 per cent higher than that of comparable DSL

packages. In sum, simulation and experiments address all three epistemological issues

discussed earlier in the paper and thus make non-metaphorical path dependence morereadily testable and falsifiable.

Third, counterfactual investigation of causal relationships offers promising methods to

study causal relationships (Durand and Vaara, 2009). Counterfactuals are conditional

statements that test logically (thought experiments) or probabilistically (computer-aided

modelling) the direction and stability of a relationship between two or more events. Using

primary and secondary data to construct event sequences, counterfactual history pro-

vides methods to ponder whether what-if questions modify the associative structure of

recorded facts (Tetlock and Belkin, 1996; Tetlock and Parker, 2006). Conceptual clarity,

cotenability, consistency, and projectability are pivotal principles that could help classical

case studies to move away from the suboptimal path on which they currently stand.

Counterfactual case studies could shed light on the contingency conditions that brought

about a triggering event and on the prevailing reinforcing mechanisms which both

combined resulted in path dependence. Causal modelling (Pearl, 2000; Spirtes,

Glymour, and Schines, 2000) offers more systematic exploration of causal relationships.

Causal modelling tests the direct and conditional associations between events, assesses

the presence of unobservable factors in a dataset, and proposes techniques to blockad

infinitum regression (e.g. conditioning, instrumenting, and mediating techniques). This

systematic exploration of data associations could prove useful in exhausting the condition

sets that prevail in defined path-dependence situations. This would avoid the ad hocspecification of A or Bs characteristics which are unverifiable or unfalsifiable or the

Path Dependence 751


17/25

CONCLUSION

Our redefinition of path dependence helps investigate the conditions under which the

concept can advance our knowledge of organizations. Path dependence, which is not

about hypersensitivity to initial conditions, can exist without increasing returns. Path

dependence predicts lock-in only when contingent events amplified by a self-reinforcement mechanism causes alternative paths to be selected out. Contingency and

various components of self-reinforcement are widespread in organizational life, so path

dependence must be at play out there, even though its most visible consequences may be

found in exceptional and long-lasting cases of extreme performance levels. So far, most

attempts to apply path dependence empirically have not satisfactorily addressed impor-

tant epistemological issues, such as how to verify or falsify contingency, how to define the

long run, and how to speak adequately of suboptimality given the core theorem of path

dependence, i.e. "(O, I), P(O|I) < 1.

This paper has explicitly opted in favour of narrowing the definition of path depen-

dence, thereby enabling scholars to specify what conditions need to be met to characterize

a historical sequence as path-dependent, andto distinguish a path-dependent process from

its outcome. In turn, we suggest more adequate methodologies to tackle the so far

overlooked issue of empirical testability. We believe a cautious and rigorous use of path

dependence has a strong potential for organization scholars interested in institutions,

technology and dynamic capabilities, provided that the concept is well defined, clearly

identifiable and distinct from sister notions, and appropriately measured. However, we

also recommend that path dependence be used parsimoniously, especially since to date,

consensual empirical evidence of path dependence is scarce. When other theoretical

notions (e.g. first mover advantage, imprinting, or inertia) and their related assumptionsdisplay a better fit with the data, there is no need to invoke path dependence (Table I).

Thus, when path dependence is used in empirical research, we recommend that

scholars: (i) stipulate what relevant properties of institutions, technologies, or capabilities

should be taken into account to compare alternative paths (in the case of VHS vs. Beta:

price, time of entry, image quality, recording time); (ii) make clear what are the poten-

tially crucial contingencies that occurred and when they occurred; and (iii) specify for

each path what components of self-reinforcement are at play and for how long. Follow-

ing these steps helps flesh out more robust hypotheses regarding the possible presence of

path dependence. Moreover, the development of controlled research designs like simu-lations, experiments, and causal modelling is the only way to potentially supply strong

evidence of this specific form of history dependence.

Our position in this paper is that exogenous shocks are required to shake the system free

of its history. This view has the merit of consistency since it preserves the explanatory

power of structures over time: for instance, we do not need to posit, for obscure reasons,

that institutions suddenly cease to produce constraining institutional effects on the

moment the system escapes lock-in (Schneiberg, 2007). However, this conception of path

dependence should be qualified as it displays some limitations. First, path self-

reinforcement is not endless. Economies of scale, for example, cannot yield additionalprofit indefinitely. As noted by Ruttan (2001), paths can sometimes exhaust themselves

endogenously. Second, paths can unfold simultaneously in several environments. Institu-



18/25

tions relate to multiple environments, firms operate in diverse industries, and dynamic

capabilities can serve different business units across markets submitted to various selection

criteria. Bassanini and Dosi (2001) show that path coevolution implies fitness maximiza-

tion across different landscapes, and as becomes visible in NK models with multiple local

peaks, misfit in one landscape can cause a major discontinuity along the whole path, as if

local lock-out created global path breaking. Therefore, there may exist endogenous

conditions for exiting the lock-in situation that will require further investigation.

Hirsch and Gillespie (2001, p. 87) ask whether path dependence contain[s] any value

added over existing theories of innovation and social changes our answer is YES.

However, path dependence is not a theoretical Russian doll hosting every possible story

of the history-matters type. There is potential for a non-metaphorical conceptualization

of path dependence to inform organizational research, just like many other theories do.

This paper has sketched the contours of a testable theory of organizational path depen-

dence, and indicated how future research could be designed. Hopefully this will help

sociologists, economists, and organization and management scholars pursue the pathdependence research agenda on a sounder basis.

ACKNOWLEDGMENTS

The authors wish to thank the participants of the Freie Universitt Berlin Conference on Path Dependence(2829 February 2008) for their feedback on an earlier draft of this paper. Comments and suggestions byFloor van den Born, Amlie Gheerbrant, Brian Hill, and Xavier Labatie are also acknowledged. Finally, theauthors would like to thank Joep Cornelissen and two reviewers for the developmental feedback they gavethroughout the review process.

NOTES

[1] Terms marked with * on their first occurrence are defined in the glossary (see Appendix 3).[2] Path dependence implies that how paths are created and how they are sustained are different mecha-

nisms (Stinchcombe, 1968, pp. 10416). Without contingency, alternative paths would be known ex anteand there would be no unpredictable outcome. So without contingency, the same structural cause wouldexplain why a path was both chosen in the first place and subsequently pursued most probably becauseit first seemed the best of all. Thus, path dependence is distinct from structural explanations to the extentthat path creation and path reinforcement are two necessary building blocks that correspond to different

underlying mechanisms (Schwartz, 2004).[3] Path dependence should clearly be distinguished from hypersensitivity to initial conditions, a notion

central to chaos theory. Path dependence deals with stochastic processes, whereas chaos theory isconcerned with deterministic yet unpredictable systems (see Table I).

[4] These examples illustrate the kind of phenomena that should attract the attention of path dependencescholars. We do not imply that they constitute indisputable evidence of path dependence, for reasonsdeveloped in the next section.

[5] In the sense described earlier in the paper. Reference to luck or chance as a source of competitiveadvantage (Barney, 1986) has triggered a comparable controversy in the field of strategic management(Priem and Butler, 2001). For theories to have empirical content, law-like necessity must characterize theassociation of two phenomena. Since luck cannot imply sustained competitive advantage, the RBVcannot uncontroversially rely on luck to predict firm success.

[6] This is a simple application of Leibnizs law of the identity of indiscernibles (Hacking, 1975; Leibniz,1686/2000), which states that no two distinct entities have got exactly the same properties (Forrest,2006; see Appendix 2.1). See Durand (2002) for a discussion of identity assumptions as a common basis

Path Dependence 753


19/25

APPENDIX 1: NON-ERGODIC MARKOV CHAINS

Path dependence is a property of non-ergodic Markov chains that have at least two

possible equilibria selected contingently along the path. A stochastic process consisting of

a sequence of random variables {Xn,n =0,1,2, . . .} with a finite or countable number of

possible values and that has, for any state i, a fixed probability Pijthat it will next be injis called aMarkov chain(e.g. Ross, 1996). Any Markov chain verifies the Markov property

stating that the future only depends on the present:

Pr , . . . , Pr .X x X x X x X x X xn n n n n n+ += = =( ) = = =( )1 1 1 1

A Markov chain is irreducibleif all states communicate with each other.

Stateihasperiodd if Piin=0 whenevernis divisible by d and d is the greatest integer

with this property. A state is said to be aperiodicwhen it has a period of 1.

State iis recurrentif with certainty a process starting at jwill eventually return; it is

positive recurrentif the expected number of transitions needed to return is non-null and

finite.

An irreducible aperiodic Markov chain with only positive recurrent states is said to

beergodic(Ross, 1996, ch. 4). It has a stationary distribution, i.e. one that verifies, for

alln:

Pr Pr .X x X y X x X yn n n n+ = =( ) = = =( )1 1

When a Markov chain is ergodic, it is possible to reach a given state from any other

state through a certain sequence of events. As a consequence, a Markov chain is

non-ergodic if it is either non-irreducible, non-aperiodic, or non-positive recurrent.

APPENDIX 2: FORMAL DEVELOPMENTS ABOUT SUBOPTIMALITY

1. Identity of the Indiscernibles

Using the notation of predicate logic, A and B representing alternative options and P any

property characterizing them, the argument about the existence of distinctive propertiescan be written:

"P[PAPB] A=B

AB ( ) ( )[ ]P PA PB v PA PB^ ^

(if for any property P, A and B have P, then A

and B are identical; A and B are distinct;

therefore, there is at least one property P along

which A and B differ.)

2. Suboptimality

The claim: we might be living in the best of all possible worlds does not contradict the



20/25

The current outcome O belongs to the set of all possible outcomes S. Assume O is the

optimal outcome (whatever the criteria for optimality). Thus O S, x S, O> x.

Let us posit the existence of Q, a better outcome than O (O < Q). Yet, we cannot

prove that Q S, since there is no deterministic relationship between the past states

and the possible outcomes. Therefore, the probability that we live in the best of all

possible worlds is non-zero.

APPENDIX 3: GLOSSARY OF TERMS MARKED WITH*

Contingency: Consistent with prior path dependence research, contingent refers to unpre-

dictable, non-purposive, and seemingly random events

Ergodicity: Property of a dynamic system wherein it is possible to reach a given state from

any other state through a certain sequence of events (more details in Appendix 1)

Falsifiability: A scientific proposition is falsifiable if it can be refuted empiricallyIncreasing (decreasing) returns: Under increasing (decreasing) returns, a choice made at time

t becomes more (less) likely to be made again by the same agent at time t+1.

Economies of scale are a particular instance of increasing returns: a firm producing

more decreases its unit cost and is therefore enticed to further increase production in

the future.

Leibniz: Leibniz (16461716) is a German philosopher, logician, and mathematician.

Lock-in: A situation of relatively stable equilibrium, caused by path dependence, from

which it is difficult to escape without the intervention of shocks exogenous to the system.

Necessary condition: A necessary condition of a proposition is one that must be satisfied for

the proposition to be true.

Positive (negative) externalities: Under positive (negative) externalities, a choice cmade at

time tby an agent increases (decreases) the utility derived by another agent from

choice c made at time t+1. Externalities differ from returns in the sense that they

modify the utility function of agents not directly involved in the initial decision (e.g.

users of a different product, future users, stakeholders).

Property: Anything that can be said to characterize an entity. Properties include the

attributes or qualities or features or characteristics of things. Properties have causal

powers.

Self-reinforcement: A set of mechanisms endogenous to a given path that makes it more and

more dominant over time relative to alternative paths.

Stochastic: A stochastic system includes a predictable and a random component. It is

non-deterministic.

Suboptimality: A situation is suboptimal if there exists a tenable alternative situation which

would increase agents utility relative to the suboptimal case.

Testability: A theory is testable when there exist methodological instruments that allow for

the investigation of the empirical propositions one derives from it.Verifiability: A proposition is verifiable if there exist some observable events pertinent for

Path Dependence 755


21/25

REFERENCES

Akkermans, H. and Romme, A. G. L. (2008). How partnership behaviour evolves in networks: pathdependency, social figuration and life events. Paper presented at the conference Studying PathDependencies of Businesses, Institutions, and Technologies, Freie Universitt Berlin, 2829 Feb 2008.

Arrow, K. (2000). Increasing returns: historiographic issues and path dependence.European Journal of the

History of Economic Thought, 7, 17180.Arthur, B. W. (1989). Competing technologies, increasing returns and lock-in by historical events.Economic

Journal, 99, 11631.Arthur, B. W. (1990). Positive feedbacks in the economy.Scientific American, 262, 929.Bach, T. (2008). Information externalities and search behaviour in a dual technology choice model the

case of German Broadband. Paper presented at the conference Studying Path Dependencies ofBusinesses, Institutions, and Technologies, Freie Universitt Berlin, 2829 Feb 2008.

Barney, J. B. (1986). Organizational culture: can it be a source of sustained competitive advantage?.Academyof Management Review, 11, 65665.

Barney, J. B. (1991). Firm resources and sustained competitive advantage.Journal of Management, 17, 99120.Bassanini, A. P. and Dosi, G. (2001). When and how chance and human will can twist the arms of Clio: an

essay on path dependence in a world of irreversibilities. In Garud, R. and Karne, P. (Eds), Path

Dependence and Creation. Mahwah, NJ: Lawrence Erlbaum, 4168.Baum, J. A. C. and Silverman, B. S. (2001). Complexity, attractors, and path dependence and creation in

technological evolution. In Garud, R. and Karne, P. (Eds),Path Dependence and Creation. Mahwah, NJ:Lawrence Erlbaum, 169209.

Becker, M., Knudsen, T. and Stieglitz, N. (2008). Managing radical and incremental change: path-dependentorganizational search in changing environments. Paper presented at the conference Studying PathDependencies of Businesses, Institutions, and Technologies, Freie Universitt Berlin, 2829 Feb 2008.

Carley, K. M. (2001). Computational approaches to sociological theorizing. In Turner, J. (Ed.), Handbookof Sociological Theory. New York: Kluwer, 6984.

Chaitin, G. J. (1975). Randomness and mathematical proof. Scientific American, 232, 4752.Cipolla, C. (1965). Guns, Sails and Empires; Technological Innovation and the Early Phases of European Expansion,

14001700. New York: Pantheon.

Cohen, W. M. and Levinthal, D. A. (1990). Absorptive capacity: a new perspective on learning andinnovation.Administrative Science Quarterly, 35, 12852.Cowan, R. (1990). Nuclear power reactors: a study in technological lock in. Journal of Economic History, 50,

54167.Cowan, R. and Gunby, P. (1996). Sprayed to death: path dependence, lock-in, and pest control strategies.

Economic Journal, 106, 52142.Cusumano, M. A., Mylonadis, Y. and Rosenbloom, R. S. (1992). Strategic maneuvering and mass-market

dynamics: the triumph of VHS over Beta. Business History Review, 66, 5194.David, P. A. (1985). Clio and the economics of QWERTY. American Economic Review Proceedings, 75, 33237.David, P. A. (1994). Why are institutions the carriers of history? Path dependence and the evolution of

conventions, organizations and institutions. Structural Change and Economic Dynamics, 5, 20520.David, P. A. (2001). Path dependence, its critics and the quest for historical economics . In Garrouste, P.

and Ioannides, S. (Eds), Evolution and Path Dependence in Economic Ideas: Past and Present. Cheltenham:Edward Elgar, 1540.Davis, J. P., Eisenhardt, K. M. and Bingham, C. B. (2007). Developing theory through simulation methods.

Academy of Management Review, 32, 48099.De Rond, M. and Thietart, R.-A. (2007). Choice, chance and inevitability in strategy. Strategic Management

Journal, 28, 53551.Dierickx, I. and Cool, K. (1989). Assets and stock accumulation and sustainability of competitive advan-

tage. Management Science, 35, 150411.Djelic, M.-L. and Quack, S. (2007). Overcoming path dependency: path generation in open systems.Theory

and Society, 36, 16186.Dooley, K. (2002). Simulation research methods. In Baum, J. A. C. (Ed.), Companion to Organizations.

London: Blackwell, 82948.Durand, R. (2002). Competitive advantages exist: a critique of Powell. Strategic Management Journal, 23,

86772.Durand, R. and Vaara, E. (2009). Causation, counterfactuals, and competitive advantage. Strategic Man-

J l 30 1245 64



22/25

Eisenhardt, K. M. and Martin, J. A. (2000). Dynamic capabilities: what are they?. Strategic ManagementJournal, 21, 110521.

Ekelund, R. B. and Tollison, R. D. (1997). On neoinstitutional theory and preclassical economies: mer-cantilism revisited. European Journal of the History of Economic Thought, 4, 37599.

Foray, D. (1997). The dynamic implications of increasing returns: technological change and path-dependent inefficiency. International Journal of Industrial Organization, 15, 73352.

Foreman-Peck, J. (1996). Technological lock-in and the power source for the motor car. Discussion Papers inEconomic and Social History, Vol. 7. Oxford: University of Oxford Press.

Forrest, P. (2006). The identity of indiscernibles. Stanford Encyclopedia of Philosophy. Available at: http://plato.stanford.edu/entries/identity-indiscernible/ (accessed 9 February 2010).

Garud, R. (2008). Conferences as venues for the configuration of emerging organizational fields: the case ofcochlear implants.Journal of Management Studies, 45, 106188.

Garud, R. and Karne, P. (Eds) (2001a). Path Dependence and Creation. Mahwah, NJ: Lawrence Erlbaum.Garud, R. and Karne, P. (2001b). Path creation as a process of mindful creation. In Garud, R. and

Karne, P. (Eds), Path Dependence and Creation. Mahwah, NJ: Lawrence Erlbaum, 140.Garud, R., Kumaraswamy, A. and Karne, P. (2010). Path dependence or path creation?. Journal of

Management Studies, 47, 76074.Gedajlovic, E., Lubatkin, M. H. and Schulze, W. S. (2004). Crossing the threshold from founder manage-

ment to professional management: a governance perspective. Journal of Management Studies, 41, 899912.

Goldstone, J. A. (1998). Initial conditions, general laws, path dependence, and explanation in historicalsociology.American Journal of Sociology, 104, 82945.

Hacking, I. (1975). The identity of indiscernibles. The Journal of Philosophy, 72, 24956.Hannan, M. T. and Freeman, J. (1984). Structural inertia and organizational change. American Sociological

Review, 49, 14964.Helfat, C. E. (1994). Evolutionary trajectories in petroleum firm R&D. Management Science, 40, 172047.Helfat, C. E. and Peteraf, M. A. (2003). The dynamic resource-based view: capability lifecycles. Strategic

Management Journal, 24, 9971010.Helfat, C. E., Finkelstein, S., Mitchell, W., Peteraf, M., Singh, H., Teece, D. J. and Winter, S. G. (2007).

Dynamic Capabilities: Understanding Strategic Change in Organizations. Oxford: Blackwell.

Hirsch, P. M. and Gillespie, J. J. (2001). Unpacking path dependence: differential valuations accordedhistory across disciplines. In Garud, R. and Karne, P. (Eds), Path Dependence and Creation. Mahwah, NJ:Lawrence Erlbaum, 6990.

Kay, A. (2005). A critique of the use of path dependency in policy studies. Public Administration, 83, 55371.Kenney, M. and von Burg, U. (2001). Paths and regions: the creation and growth of Silicon Valley.

In Garud, R. and Karne, P. (Eds), Path Dependence and Creation. Mahwah, NJ: Lawrence Erlbaum,12747.

Koch, J., Eisend, M. and Petermann, A. (2008). The impact of complexity on path-dependent decision-making processes: an experimental study. Paper presented at the conference Studying Path Depen-dencies of Businesses, Institutions, and Technologies. Freie Universitt Berlin, 2829 Feb 2008.

Kogut, B. and Zander, U. (1992). Knowledge of the firm, combinative capabilities, and the replication oftechnology.Organization Science, 3, 38397.

Lant, T. K. and Mezias, S. J. (1990). Managing discontinuous change: a simulation study of organizationallearning and entrepreneurship. Strategic Management Journal, 11, 14779.Leibniz, G. W. (2000)[1686]. Discours de Mtaphysique. Paris: Vrin.Leonard-Barton, D. (1992). Core capabilities and core rigidities: a paradox in managing new product

development.Strategic Management Journal, 13, 11125.Li, M. and Vitanyi, P. (1997).An Introduction to Kolmogorov Complexity and its Applications, 2nd edition. New York:

Springer Verlag.Lieberman, M. B. and Montgomery, D. B. (1988). First-mover advantages.Strategic Management Journal, 9,

4158.Liebowitz, S. J. and Margolis, S. E. (1990). The fable of the keys. Journal of Law & Economics, 33, 127.Liebowitz, S. J. and Margolis, S. N. (1994). Network externality: an uncommon tragedy. Journal of Economic

Perspectives, 8, 13350.Liebowitz, S. J. and Margolis, S. E. (1995). Path dependence, lock-in, and history. Journal of Law, Economics

& Organization, 11, 20526.Mahoney, J. (2000). Path dependence in historical sociology. Theory and Society, 29, 50748.M h J d Ol J (1989) R di i I tit ti Th O i ti l B i f P liti N Y k F P

Path Dependence 757


23/25

Marquis, C. (2003). The pressure of the past: network imprinting in intercorporate communities. Admin-istrative Science Quarterly, 48, 65589.

McKelvey, B. (1999). Avoiding complexity catastrophe in coevolutionary pockets: strategies for ruggedlandscapes.Organization Science, 10, 294321.

Morgan, G. and Kubo, I. (2005). Beyond path dependency? Constructing new models for institutionalchange: the case of capital markets in Japan. Socio-Economic Review, 3, 5582.

North, D. C. (1990).Institutions, Institutional Change and Economic Performance. Cambridge: Cambridge Univer-sity Press.

North, D. C. (1993). The New Institutional Economics and Development. Working Paper. St Louis: WashingtonUniversity.

Page, S. E. (2006). Path dependence. Quarterly Journal of Political Science, 1, 87115.Pearl, J. (2000).Causality: Models, Reasoning, and Inference. Cambridge, UK: Cambridge University Press.Pierson, P. (2000). Increasing returns, path dependence, and the study of politics. American Political Science

Review, 92, 25167.Porac, J. F., Rosa, J. A., Spanjol, J. and Saxon, M. S. (2001). Americas family vehicle: path creation in the

U.S. minivan market. In Garud, R. and Karne, P. (Eds), Path Dependence and Creation. Mahwah, NJ:Lawrence Erlbaum, 21342.

Priem, R. L. and Butler, J. E. (2001). Is the resource-based view a useful perspective for strategic

management research?.Academy of Management Review, 26, 2240.Rao, H. and Singh, J. (2001). The construction of new paths: institution-building activity in the early

automobile and biotechnology industries. In Garud, R. and Karne, P. (Eds), Path Dependence andCreation. Mahwah, NJ: Lawrence Erlbaum, 24367.

Ross, S. M. (1996).Stochastic Processes, 2nd edition. Chichester: John Wiley & Sons.Ruttan, V. W. (2001). Sources of technical change: induced innovation, evolutionary theory, and path

dependence. In Garud, R. and Karne, P. (Eds), Path Dependence and Creation. Mahwah, NJ: LawrenceErlbaum, 91123.

Schneiberg, M. (2007). Whats on the path? Path dependence, organizational diversity and the problem ofinstitutional change in the US economy, 19001950. Socio-Economic Review, 5, 4780.

Schreygg, G. and Blinn, M. (2008). Dubbing forever? Path dependency of the German film industrythrough historical events. Paper presented at the 2008 AOM Conference in Anaheim, California.

Schreygg, G. and Kliesch-Eberl, M. (2007). How dynamic can organizational capabilities be? Towards adual-process model of capability dynamization. Strategic Management Journal, 28, 91333.Schwartz, H. (2004). Down the Wrong Path: Path Dependence, Increasing Returns, and Historical Institutionalism .

Mimeo. Department of Politics: University of Virginia.Scott, W. R. (1991). Unpacking institutional arguments. In DiMaggio, P. J. and Powell, W. W. (Eds), The

New Institutionalism in Organizational Analysis. Chicago, IL: University of Chicago Press, 16482.Scott, W. R. (1995). Institutions and Organizations. Thousand Oaks, CA: Sage.Selznick, P. (1949). TVA and the Grass Roots. Berkeley, CA: University of California Press.Selznick, P. (1957). Leadership in Administration. New York: Harper & Row.Solomonoff, R. J. (1960).A Preliminary Report on a General Theory of Inductive Inference. Report V-131. Cambridge,

MA: Zator.Spirtes, P., Glymour, C. and Scheines, R. (2000). Causation, Prediction and Search, 2nd edition. Cambridge,

MA: MIT Press.Sterman, J. D. and Wittenberg, J. (1999). Path-dependence, competition, and succession in the dynamics ofscientific revolution.Organization Science, 10, 32241.

Stinchcombe, A. L. (1965). Social structure and organizations. In March, J. G. (Ed.), Handbook of Organi-zations. Chicago, IL: Rand McNally, 14293.

Stinchcombe, A. L. (1968). Constructing Social Theories. Chicago, IL: University of Chicago Press.Sydow, J., Schreygg, G. and Koch, J. (2009). Organizational path dependence: opening the black box.

Academy of Management Review, 34, 689709.Teece, D. J. (2007). Explicating dynamic capabilities: the nature and microfoundations of (sustainable)

enterprise performance.Strategic Management Journal, 28, 131950.Teece, D. J., Rumelt, R., Dosi, G. and Winter, S. (1994). Understanding corporate coherence theory and

evidence.Journal of Economic Behavior & Organization, 23, 130.Teece, D. J., Pisano, G. and Shuen, A. (1997). Dynamic capabilities and strategic management. Strategic

Management Journal, 18, 50935.Tetlock, P. E. and Belkin, A. (1996). Counterfactual Thought Experiments in World Politics: Logical, Methodological,

d P h l i l P p ti P i t P i t U i it P



24/25

Tetlock, P. E. and Parker, G. (2006). Counterfactual thought experiments: why we cant live with them andhow we must learn to live with them. In Tetlock, P. E., Lebow, R. N. and Parker, G. (Eds), Unmakingthe West: What-If Scenarios that Rewrite World History. New York: Cambridge University Press.

Thelen, K. (1999). Historical institutionalism in comparative politics. Annual Review of Political Science, 2,369404.

Tripsas, M. (1997). Unaveling the process of creative destruction: complementary assets and incumbent

survival in the typesetter industry. Strategic Management Journal, 18, 11942.Vergne, J. P. and Durand, R. (2010). The path of most persistence: an evolutionary perspective on path

dependence and dynamic capabilities.Organization Studies, forthcoming.Webster, M., Jr and Sell, J. (2007). Laboratory Experiments in the Social Sciences. San Diego, CA: Academic Press.Williamson, O. E. (1999). Strategy research: governance and competence perspectives. Strategic Management

Journal, 20, 1087108.Winter, S. G., Cattani, G. and Dorsch, A. (2007). The value of moderate obsession: insights from a new

model of organizational search.Organization Science, 18, 40319.Zott, C. (2003). Dynamic capabilities and the emergence of intraindustry differential firm performance:

insights from a simulation study. Strategic Management Journal, 24, 97125.

Path Dependence 759


25/25

Copyright of Journal of Management Studies is the property of Wiley-Blackwell and its content may not be

copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written

permission. However, users may print, download, or email articles for individual use.

jean-philippe vergne y rodolphe durand - theory and empirics path dependence

Documents